4 # Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc.
6 # This program is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2 of the License, or
9 # (at your option) any later version.
11 # This program is distributed in the hope that it will be useful, but
12 # WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 # General Public License for more details.
16 # You should have received a copy of the GNU General Public License
17 # along with this program; if not, write to the Free Software
18 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 """Inter-node RPC library.
26 # pylint: disable=C0103,R0201,R0904
27 # C0103: Invalid name, since call_ are not valid
28 # R0201: Method could be a function, we keep all rpcs instance methods
29 # as not to change them back and forth between static/instance methods
30 # if they need to start using instance attributes
31 # R0904: Too many public methods
40 from ganeti import utils
41 from ganeti import objects
42 from ganeti import http
43 from ganeti import serializer
44 from ganeti import constants
45 from ganeti import errors
46 from ganeti import netutils
47 from ganeti import ssconf
48 from ganeti import runtime
49 from ganeti import compat
50 from ganeti import rpc_defs
52 # Special module generated at build time
53 from ganeti import _generated_rpc
55 # pylint has a bug here, doesn't see this import
56 import ganeti.http.client # pylint: disable=W0611
59 # Timeout for connecting to nodes (seconds)
60 _RPC_CONNECT_TIMEOUT = 5
62 _RPC_CLIENT_HEADERS = [
63 "Content-type: %s" % http.HTTP_APP_JSON,
67 # Various time constants for the timeout table
68 _TMO_URGENT = 60 # one minute
69 _TMO_FAST = 5 * 60 # five minutes
70 _TMO_NORMAL = 15 * 60 # 15 minutes
71 _TMO_SLOW = 3600 # one hour
75 #: Special value to describe an offline host
80 """Initializes the module-global HTTP client manager.
82 Must be called before using any RPC function and while exactly one thread is
86 # curl_global_init(3) and curl_global_cleanup(3) must be called with only
87 # one thread running. This check is just a safety measure -- it doesn't
89 assert threading.activeCount() == 1, \
90 "Found more than one active thread when initializing pycURL"
92 logging.info("Using PycURL %s", pycurl.version)
94 pycurl.global_init(pycurl.GLOBAL_ALL)
98 """Stops the module-global HTTP client manager.
100 Must be called before quitting the program and while exactly one thread is
104 pycurl.global_cleanup()
107 def _ConfigRpcCurl(curl):
108 noded_cert = str(constants.NODED_CERT_FILE)
110 curl.setopt(pycurl.FOLLOWLOCATION, False)
111 curl.setopt(pycurl.CAINFO, noded_cert)
112 curl.setopt(pycurl.SSL_VERIFYHOST, 0)
113 curl.setopt(pycurl.SSL_VERIFYPEER, True)
114 curl.setopt(pycurl.SSLCERTTYPE, "PEM")
115 curl.setopt(pycurl.SSLCERT, noded_cert)
116 curl.setopt(pycurl.SSLKEYTYPE, "PEM")
117 curl.setopt(pycurl.SSLKEY, noded_cert)
118 curl.setopt(pycurl.CONNECTTIMEOUT, _RPC_CONNECT_TIMEOUT)
122 """RPC-wrapper decorator.
124 When applied to a function, it runs it with the RPC system
125 initialized, and it shutsdown the system afterwards. This means the
126 function must be called without RPC being initialized.
129 def wrapper(*args, **kwargs):
132 return fn(*args, **kwargs)
139 """Compresses a string for transport over RPC.
141 Small amounts of data are not compressed.
146 @return: Encoded data to send
149 # Small amounts of data are not compressed
151 return (constants.RPC_ENCODING_NONE, data)
153 # Compress with zlib and encode in base64
154 return (constants.RPC_ENCODING_ZLIB_BASE64,
155 base64.b64encode(zlib.compress(data, 3)))
158 class RpcResult(object):
161 This class holds an RPC result. It is needed since in multi-node
162 calls we can't raise an exception just because one one out of many
163 failed, and therefore we use this class to encapsulate the result.
165 @ivar data: the data payload, for successful results, or None
166 @ivar call: the name of the RPC call
167 @ivar node: the name of the node to which we made the call
168 @ivar offline: whether the operation failed because the node was
169 offline, as opposed to actual failure; offline=True will always
170 imply failed=True, in order to allow simpler checking if
171 the user doesn't care about the exact failure mode
172 @ivar fail_msg: the error message if the call failed
175 def __init__(self, data=None, failed=False, offline=False,
176 call=None, node=None):
177 self.offline = offline
182 self.fail_msg = "Node is marked offline"
183 self.data = self.payload = None
185 self.fail_msg = self._EnsureErr(data)
186 self.data = self.payload = None
189 if not isinstance(self.data, (tuple, list)):
190 self.fail_msg = ("RPC layer error: invalid result type (%s)" %
194 self.fail_msg = ("RPC layer error: invalid result length (%d), "
195 "expected 2" % len(self.data))
197 elif not self.data[0]:
198 self.fail_msg = self._EnsureErr(self.data[1])
203 self.payload = data[1]
205 for attr_name in ["call", "data", "fail_msg",
206 "node", "offline", "payload"]:
207 assert hasattr(self, attr_name), "Missing attribute %s" % attr_name
211 """Helper to ensure we return a 'True' value for error."""
215 return "No error information"
217 def Raise(self, msg, prereq=False, ecode=None):
218 """If the result has failed, raise an OpExecError.
220 This is used so that LU code doesn't have to check for each
221 result, but instead can call this function.
224 if not self.fail_msg:
227 if not msg: # one could pass None for default message
228 msg = ("Call '%s' to node '%s' has failed: %s" %
229 (self.call, self.node, self.fail_msg))
231 msg = "%s: %s" % (msg, self.fail_msg)
233 ec = errors.OpPrereqError
235 ec = errors.OpExecError
236 if ecode is not None:
240 raise ec(*args) # pylint: disable=W0142
243 def _SsconfResolver(node_list,
244 ssc=ssconf.SimpleStore,
245 nslookup_fn=netutils.Hostname.GetIP):
246 """Return addresses for given node names.
248 @type node_list: list
249 @param node_list: List of node names
251 @param ssc: SimpleStore class that is used to obtain node->ip mappings
252 @type nslookup_fn: callable
253 @param nslookup_fn: function use to do NS lookup
254 @rtype: list of tuple; (string, string)
255 @return: List of tuples containing node name and IP address
259 iplist = ss.GetNodePrimaryIPList()
260 family = ss.GetPrimaryIPFamily()
261 ipmap = dict(entry.split() for entry in iplist)
264 for node in node_list:
267 ip = nslookup_fn(node, family=family)
268 result.append((node, ip))
273 class _StaticResolver:
274 def __init__(self, addresses):
275 """Initializes this class.
278 self._addresses = addresses
280 def __call__(self, hosts):
281 """Returns static addresses for hosts.
284 assert len(hosts) == len(self._addresses)
285 return zip(hosts, self._addresses)
288 def _CheckConfigNode(name, node):
289 """Checks if a node is online.
292 @param name: Node name
293 @type node: L{objects.Node} or None
294 @param node: Node object
298 # Depend on DNS for name resolution
307 def _NodeConfigResolver(single_node_fn, all_nodes_fn, hosts):
308 """Calculate node addresses using configuration.
311 # Special case for single-host lookups
314 return [_CheckConfigNode(name, single_node_fn(name))]
316 all_nodes = all_nodes_fn()
317 return [_CheckConfigNode(name, all_nodes.get(name, None))
322 def __init__(self, resolver, port, lock_monitor_cb=None):
323 """Initializes this class.
325 @param resolver: callable accepting a list of hostnames, returning a list
326 of tuples containing name and IP address (IP address can be the name or
327 the special value L{_OFFLINE} to mark offline machines)
329 @param port: TCP port
330 @param lock_monitor_cb: Callable for registering with lock monitor
333 self._resolver = resolver
335 self._lock_monitor_cb = lock_monitor_cb
338 def _PrepareRequests(hosts, port, procedure, body, read_timeout):
339 """Prepares requests by sorting offline hosts into separate list.
342 @param body: a dictionary with per-host body data
348 assert isinstance(body, dict)
349 assert len(body) == len(hosts)
350 assert compat.all(isinstance(v, str) for v in body.values())
351 assert frozenset(map(compat.fst, hosts)) == frozenset(body.keys()), \
352 "%s != %s" % (hosts, body.keys())
354 for (name, ip) in hosts:
356 # Node is marked as offline
357 results[name] = RpcResult(node=name, offline=True, call=procedure)
360 http.client.HttpClientRequest(str(ip), port,
361 http.HTTP_PUT, str("/%s" % procedure),
362 headers=_RPC_CLIENT_HEADERS,
363 post_data=body[name],
364 read_timeout=read_timeout,
365 nicename="%s/%s" % (name, procedure),
366 curl_config_fn=_ConfigRpcCurl)
368 return (results, requests)
371 def _CombineResults(results, requests, procedure):
372 """Combines pre-computed results for offline hosts with actual call results.
375 for name, req in requests.items():
376 if req.success and req.resp_status_code == http.HTTP_OK:
377 host_result = RpcResult(data=serializer.LoadJson(req.resp_body),
378 node=name, call=procedure)
380 # TODO: Better error reporting
386 logging.error("RPC error in %s on node %s: %s", procedure, name, msg)
387 host_result = RpcResult(data=msg, failed=True, node=name,
390 results[name] = host_result
394 def __call__(self, hosts, procedure, body, read_timeout=None,
395 _req_process_fn=http.client.ProcessRequests):
396 """Makes an RPC request to a number of nodes.
398 @type hosts: sequence
399 @param hosts: Hostnames
400 @type procedure: string
401 @param procedure: Request path
402 @type body: dictionary
403 @param body: dictionary with request bodies per host
404 @type read_timeout: int or None
405 @param read_timeout: Read timeout for request
408 assert read_timeout is not None, \
409 "Missing RPC read timeout for procedure '%s'" % procedure
411 (results, requests) = \
412 self._PrepareRequests(self._resolver(hosts), self._port, procedure,
415 _req_process_fn(requests.values(), lock_monitor_cb=self._lock_monitor_cb)
417 assert not frozenset(results).intersection(requests)
419 return self._CombineResults(results, requests, procedure)
422 class _RpcClientBase:
423 def __init__(self, resolver, encoder_fn, lock_monitor_cb=None):
424 """Initializes this class.
427 self._proc = _RpcProcessor(resolver,
428 netutils.GetDaemonPort(constants.NODED),
429 lock_monitor_cb=lock_monitor_cb)
430 self._encoder = compat.partial(self._EncodeArg, encoder_fn)
433 def _EncodeArg(encoder_fn, (argkind, value)):
440 return encoder_fn(argkind)(value)
442 def _Call(self, cdef, node_list, args):
443 """Entry point for automatically generated RPC wrappers.
446 (procedure, _, timeout, argdefs, prep_fn, postproc_fn, _) = cdef
448 if callable(timeout):
449 read_timeout = timeout(args)
451 read_timeout = timeout
453 enc_args = map(self._encoder, zip(map(compat.snd, argdefs), args))
455 # for a no-op prep_fn, we serialise the body once, and then we
456 # reuse it in the dictionary values
457 body = serializer.DumpJson(enc_args)
458 pnbody = dict((n, body) for n in node_list)
460 # for a custom prep_fn, we pass the encoded arguments and the
461 # node name to the prep_fn, and we serialise its return value
462 assert(callable(prep_fn))
463 pnbody = dict((n, serializer.DumpJson(prep_fn(n, enc_args)))
466 result = self._proc(node_list, procedure, pnbody,
467 read_timeout=read_timeout)
470 return dict(map(lambda (key, value): (key, postproc_fn(value)),
476 def _ObjectToDict(value):
477 """Converts an object to a dictionary.
479 @note: See L{objects}.
482 return value.ToDict()
485 def _ObjectListToDict(value):
486 """Converts a list of L{objects} to dictionaries.
489 return map(_ObjectToDict, value)
492 def _EncodeNodeToDiskDict(value):
493 """Encodes a dictionary with node name as key and disk objects as values.
496 return dict((name, _ObjectListToDict(disks))
497 for name, disks in value.items())
500 def _PrepareFileUpload(filename):
501 """Loads a file and prepares it for an upload to nodes.
504 data = _Compress(utils.ReadFile(filename))
505 st = os.stat(filename)
506 getents = runtime.GetEnts()
507 return [filename, data, st.st_mode, getents.LookupUid(st.st_uid),
508 getents.LookupGid(st.st_gid), st.st_atime, st.st_mtime]
511 def _PrepareFinalizeExportDisks(snap_disks):
512 """Encodes disks for finalizing export.
517 for disk in snap_disks:
518 if isinstance(disk, bool):
519 flat_disks.append(disk)
521 flat_disks.append(disk.ToDict())
526 def _EncodeImportExportIO((ieio, ieioargs)):
527 """Encodes import/export I/O information.
530 if ieio == constants.IEIO_RAW_DISK:
531 assert len(ieioargs) == 1
532 return (ieio, (ieioargs[0].ToDict(), ))
534 if ieio == constants.IEIO_SCRIPT:
535 assert len(ieioargs) == 2
536 return (ieio, (ieioargs[0].ToDict(), ieioargs[1]))
538 return (ieio, ieioargs)
541 def _EncodeBlockdevRename(value):
542 """Encodes information for renaming block devices.
545 return [(d.ToDict(), uid) for d, uid in value]
550 rpc_defs.ED_OBJECT_DICT: _ObjectToDict,
551 rpc_defs.ED_OBJECT_DICT_LIST: _ObjectListToDict,
552 rpc_defs.ED_NODE_TO_DISK_DICT: _EncodeNodeToDiskDict,
553 rpc_defs.ED_FILE_DETAILS: _PrepareFileUpload,
554 rpc_defs.ED_COMPRESS: _Compress,
555 rpc_defs.ED_FINALIZE_EXPORT_DISKS: _PrepareFinalizeExportDisks,
556 rpc_defs.ED_IMPEXP_IO: _EncodeImportExportIO,
557 rpc_defs.ED_BLOCKDEV_RENAME: _EncodeBlockdevRename,
561 class RpcRunner(_RpcClientBase,
562 _generated_rpc.RpcClientDefault,
563 _generated_rpc.RpcClientBootstrap,
564 _generated_rpc.RpcClientConfig):
568 def __init__(self, context):
569 """Initialized the RPC runner.
571 @type context: C{masterd.GanetiContext}
572 @param context: Ganeti context
575 self._cfg = context.cfg
577 encoders = _ENCODERS.copy()
579 # Add encoders requiring configuration object
581 rpc_defs.ED_INST_DICT: self._InstDict,
582 rpc_defs.ED_INST_DICT_HVP_BEP: self._InstDictHvpBep,
583 rpc_defs.ED_INST_DICT_OSP: self._InstDictOsp,
586 # Resolver using configuration
587 resolver = compat.partial(_NodeConfigResolver, self._cfg.GetNodeInfo,
588 self._cfg.GetAllNodesInfo)
590 # Pylint doesn't recognize multiple inheritance properly, see
591 # <http://www.logilab.org/ticket/36586> and
592 # <http://www.logilab.org/ticket/35642>
593 # pylint: disable=W0233
594 _RpcClientBase.__init__(self, resolver, encoders.get,
595 lock_monitor_cb=context.glm.AddToLockMonitor)
596 _generated_rpc.RpcClientConfig.__init__(self)
597 _generated_rpc.RpcClientBootstrap.__init__(self)
598 _generated_rpc.RpcClientDefault.__init__(self)
600 def _InstDict(self, instance, hvp=None, bep=None, osp=None):
601 """Convert the given instance to a dict.
603 This is done via the instance's ToDict() method and additionally
604 we fill the hvparams with the cluster defaults.
606 @type instance: L{objects.Instance}
607 @param instance: an Instance object
608 @type hvp: dict or None
609 @param hvp: a dictionary with overridden hypervisor parameters
610 @type bep: dict or None
611 @param bep: a dictionary with overridden backend parameters
612 @type osp: dict or None
613 @param osp: a dictionary with overridden os parameters
615 @return: the instance dict, with the hvparams filled with the
619 idict = instance.ToDict()
620 cluster = self._cfg.GetClusterInfo()
621 idict["hvparams"] = cluster.FillHV(instance)
623 idict["hvparams"].update(hvp)
624 idict["beparams"] = cluster.FillBE(instance)
626 idict["beparams"].update(bep)
627 idict["osparams"] = cluster.SimpleFillOS(instance.os, instance.osparams)
629 idict["osparams"].update(osp)
630 for nic in idict["nics"]:
631 nic['nicparams'] = objects.FillDict(
632 cluster.nicparams[constants.PP_DEFAULT],
636 def _InstDictHvpBep(self, (instance, hvp, bep)):
637 """Wrapper for L{_InstDict}.
640 return self._InstDict(instance, hvp=hvp, bep=bep)
642 def _InstDictOsp(self, (instance, osparams)):
643 """Wrapper for L{_InstDict}.
646 return self._InstDict(instance, osp=osparams)
649 class JobQueueRunner(_RpcClientBase, _generated_rpc.RpcClientJobQueue):
650 """RPC wrappers for job queue.
653 def __init__(self, context, address_list):
654 """Initializes this class.
657 if address_list is None:
658 resolver = _SsconfResolver
660 # Caller provided an address list
661 resolver = _StaticResolver(address_list)
663 _RpcClientBase.__init__(self, resolver, _ENCODERS.get,
664 lock_monitor_cb=context.glm.AddToLockMonitor)
665 _generated_rpc.RpcClientJobQueue.__init__(self)
668 class BootstrapRunner(_RpcClientBase, _generated_rpc.RpcClientBootstrap):
669 """RPC wrappers for bootstrapping.
673 """Initializes this class.
676 _RpcClientBase.__init__(self, _SsconfResolver, _ENCODERS.get)
677 _generated_rpc.RpcClientBootstrap.__init__(self)
680 class ConfigRunner(_RpcClientBase, _generated_rpc.RpcClientConfig):
681 """RPC wrappers for L{config}.
684 def __init__(self, context, address_list):
685 """Initializes this class.
689 lock_monitor_cb = context.glm.AddToLockMonitor
691 lock_monitor_cb = None
693 if address_list is None:
694 resolver = _SsconfResolver
696 # Caller provided an address list
697 resolver = _StaticResolver(address_list)
699 _RpcClientBase.__init__(self, resolver, _ENCODERS.get,
700 lock_monitor_cb=lock_monitor_cb)
701 _generated_rpc.RpcClientConfig.__init__(self)